diff --git a/arch/powerpc/kvm/book3s.c b/arch/powerpc/kvm/book3s.c
index c8ead7b9f03b..3f2a8360c857 100644
--- a/arch/powerpc/kvm/book3s.c
+++ b/arch/powerpc/kvm/book3s.c
@@ -291,6 +291,9 @@ pfn_t kvmppc_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 {
 	ulong mp_pa = vcpu->arch.magic_page_pa;
 
+	if (!(vcpu->arch.shared->msr & MSR_SF))
+		mp_pa = (uint32_t)mp_pa;
+
 	/* Magic page override */
 	if (unlikely(mp_pa) &&
 	    unlikely(((gfn << PAGE_SHIFT) & KVM_PAM) ==
diff --git a/arch/powerpc/kvm/book3s_pr.c b/arch/powerpc/kvm/book3s_pr.c
index 912e10fbf6ef..dba282e5093f 100644
--- a/arch/powerpc/kvm/book3s_pr.c
+++ b/arch/powerpc/kvm/book3s_pr.c
@@ -145,6 +145,21 @@ void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 		}
 	}
 
+	/*
+	 * When switching from 32 to 64-bit, we may have a stale 32-bit
+	 * magic page around, we need to flush it. Typically 32-bit magic
+	 * page will be instanciated when calling into RTAS. Note: We
+	 * assume that such transition only happens while in kernel mode,
+	 * ie, we never transition from user 32-bit to kernel 64-bit with
+	 * a 32-bit magic page around.
+	 */
+	if (vcpu->arch.magic_page_pa &&
+	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
+		/* going from RTAS to normal kernel code */
+		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
+				     ~0xFFFUL);
+	}
+
 	/* Preload FPU if it's enabled */
 	if (vcpu->arch.shared->msr & MSR_FP)
 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
@@ -252,6 +267,9 @@ static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 {
 	ulong mp_pa = vcpu->arch.magic_page_pa;
 
+	if (!(vcpu->arch.shared->msr & MSR_SF))
+		mp_pa = (uint32_t)mp_pa;
+
 	if (unlikely(mp_pa) &&
 	    unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 		return 1;